MPs have called for legislation imposing enforceable safety obligations on developers of advanced artificial intelligence, warning that the UK’s existing system leaves major decisions about risk largely dependent on the companies building the technology.
The House of Commons Business, Innovation, Science and Trade Committee has identified ten gaps in the oversight of frontier AI and recommended an AI Safety and Security Bill to address them.
Its report, Guardrails for growth: AI safety and security, calls for independent scrutiny of high-risk systems, mandatory reporting of significant incidents, stronger whistleblower protection, and powers to restrict or suspend unsafe development and deployment.
The committee argues that developers should not have sole responsibility for establishing whether the risks associated with increasingly capable AI models are acceptable.
It wants ministers, accountable to Parliament, to determine the relevant safety thresholds, supported by independent technical assessments and regulatory enforcement.
The proposals follow evidence sessions involving AI researchers, former industry employees, and senior representatives of OpenAI, Google DeepMind, Anthropic, and Meta.
The committee’s findings are recommendations to government. They do not create new statutory obligations, and the proposed Bill has not been enacted.
Independent testing and incident reporting —
One concern identified by the committee is the absence of mandatory, independently scrutinised assessments before designated high-risk AI systems are deployed.
Developers can currently conduct internal safety testing and cooperate voluntarily with external evaluators, but the committee argues that these arrangements do not provide sufficiently dependable oversight of the most capable systems.
It recommends requiring developers to substantiate safety claims and give independent evaluators meaningful access to relevant models and supporting information.
Material changes in capabilities, safeguards, or intended uses could trigger further assessments under the proposed framework.
Where risks cannot be controlled adequately, regulators would need authority to restrict or suspend development or deployment.
The committee also calls for mandatory reporting of serious incidents and near misses. Such information could help public authorities identify recurring problems that might otherwise remain within individual technology companies.
Protected reporting channels for whistleblowers form another recommendation, reflecting concerns that employees may encounter evidence of serious risks before those issues become apparent outside an organisation.
The report identifies particular challenges associated with open-weight AI models, where releasing model components can make subsequent restrictions or withdrawal difficult.
Additional safeguards would therefore need to reflect the potential irreversibility of some releases rather than rely solely on intervention after deployment.
Clarifying responsibility across AI systems —
The committee wants clearer arrangements for assigning responsibility when AI causes harm.
Advanced systems can involve model developers, infrastructure suppliers, application providers, and organisations deploying the technology. Their respective responsibilities may differ according to the capabilities they control and the decisions they make.
MPs recommend stronger routes to redress, appropriate criminal offences for harmful misuse, and a broader digital rights framework.
These proposals would affect developers most directly, while potentially changing the information and assurances available to businesses purchasing or deploying advanced AI systems.
Independent testing could also create opportunities for companies specialising in AI evaluation, assurance, and governance.
The committee regards these capabilities as a potential UK export industry, building on the country’s existing AI safety research and testing expertise.
It uses OECD productivity estimates to illustrate an economic scenario in which AI could leave Britain’s economy up to £400bn larger over ten years. The figure is conditional and should not be treated as a guaranteed economic benefit.
Alongside domestic regulation, the committee wants ministers to use the UK’s 2027 G20 presidency to pursue international cooperation on model access, safety assessments, incident reporting, and high-risk releases.
It has requested clear negotiating objectives and milestones, while emphasising that international discussions should not delay domestic safeguards.
The government is expected to respond to the parliamentary report. The resulting policy decisions will determine whether the proposed statutory framework is developed and which obligations would apply to advanced AI developers.




